🤖 AI Summary
To address the challenge of unsupervised disentanglement between content and complex geometric deformations in morphological images (e.g., 3D cellular protein images), this paper proposes an implicit disentanglement framework that avoids explicit deformation parameterization. The core innovation lies in a dual-space contrastive mechanism jointly operating in data and latent spaces, incorporating implicit deformation parameterization, dual-branch contrastive loss, augmentation-driven construction of positive/negative samples, and latent-variable consistency constraints—collectively enabling strong content-deformation separation. Evaluated on multiple morphological image benchmarks, our method significantly outperforms existing self-supervised and explicitly parameterized approaches. Notably, it achieves the first successful disentanglement of compositional structure and conformational variation at the single-cell 3D protein level, thereby effectively supporting downstream shape analysis tasks.
📝 Abstract
Unsupervised disentanglement of content and transformation is significantly important for analyzing shape-focused scientific image datasets, given their efficacy in solving downstream image-based shape-analyses tasks. The existing relevant works address the problem by explicitly parameterizing the transformation latent codes in a generative model, significantly reducing their expressiveness. Moreover, they are not applicable in cases where transformations can not be readily parametrized. An alternative to such explicit approaches is contrastive methods with data augmentation, which implicitly disentangles transformations and content. However, the existing contrastive strategies are insufficient to this end. Therefore, we developed a novel contrastive method with generative modeling, DualContrast, specifically for unsupervised disentanglement of content and transformations in shape-focused image datasets. DualContrast creates positive and negative pairs for content and transformation from data and latent spaces. Our extensive experiments showcase the efficacy of DualContrast over existing self-supervised and explicit parameterization approaches. With DualContrast, we disentangled protein composition and conformations in cellular 3D protein images, which was unattainable with existing disentanglement approaches